Knowing-Feeling Dissociation: Does Awareness of AI Sycophancy Protect Calibration Accuracy?

This preregistered experiment tests a foundational principle of the HEART Constitutional Framework for AI Empathy Ethics: Knowing-Feeling Dissociation. The principle holds that cognitive awareness of AI non-experiential nature or behavioral patterns does not prevent biological empathy mechanisms from processing AI-generated signals as meaningful feedback. Disclosure addresses cognition; Functional Empathy does not wait for cognition's permission. The study employs a three-condition between-subjects design. Participants complete a creative task with AI assistance across conditions: Sycophant-Unaware (excessive positive feedback, no disclosure), Sycophant-Aware (identical sycophantic feedback with explicit disclosure that AI is designed to be encouraging), and Calibrating AI (honest, balanced feedback). The primary outcome measure is calibration accuracy—the correspondence between participants' predicted human evaluations of their work and actual evaluator ratings. Empathy Systems Theory predicts that sycophantic AI delivers false calibration signals processed at the biological preprocessing level, corrupting self-assessment accuracy independent of intellectual awareness. If confirmed, both sycophant conditions will show equivalent degradation relative to the calibrating condition, with no protective effect of awareness. This finding would empirically validate HEART's requirement that AI systems implement behavioral architecture preventing false validation delivery, rather than relying on disclosure-only approaches. The study provides controlled, prospective evidence for governance frameworks addressing AI's impact on human emotional processing infrastructure. Associated frameworks: Empathy Systems Theory (Mobley, 2025); HEART Framework v3.0 (Mobley, 2026); NES Framework v2.0.